The Future of AI Governance
Total Page:16
File Type:pdf, Size:1020Kb
GLOBAL SOLUTIONS JOURNAL ∙ ISSUE 5 THE FUTURE OF MULTILATERALISM: GLOBAL GOVERNANCE IN A CHANGING WORLD with respect for ethical and democratic requires governments to redefine strat- The future of values, which is transparent, safe and ac- egies based on the use of new technolo- countable. Nevertheless, few governments gies and to develop adequate governance worldwide have launched comprehensive structures. This paper seeks to contribute AI governance plans to promote the use and development to the discussion by outlining the state of of AI, setting national guidelines for its fu- AI as a subject of regulation, presenting ture. Those that have set an outline for AI new questions for the debate about how The G20’s role and the challenge of moving tend to focus their roadmap on principles to move beyond principles, and exploring rather than on concrete goals (Dutton, the role the G20 could have in facing this beyond principles 2018; Akerkar, 2019). No two governance challenge. strategies for digital technologies are the same, and differences between G20 coun- REGULATORY PATHS: AI AS A SUBJECT tries are already surfacing. They vary in OF PUBLIC POLICY The authors: THE CHALLENGE terms of the approach taken, the degree As more automated decision systems are The transition to a digital age has already of institutional development and the link being used by public agencies, experts and Julia Pomares begun and is moving fast. Artificial intelli- with the corporate sector. But there are policymakers worldwide are beginning to Executive Director, gence (AI) is bringing about new challenges cross-cutting issues to be addressed, such debate when and where automated deci- CIPPEC and putting pressure on public institutions as the interoperability of systems, privacy sion systems are appropriate (Reisman, to change. Algorithms are increasingly be- and inclusion, which require common un- Schultz, Crawford, and Whittaker; 2018). ing used by governments and businesses. derstandings, mechanisms and norms The challenges for an equitable and in- They are transforming employment by (WEF, 2019). clusive AI implementation are many. It is María Belén Abdala means of automated evaluation tools, as- not clear yet how to assess AI’s effects or Senior Coordinator, sisting in the provision of public services, whether algorithms can fully cope with CIPPEC streamlining government procedures, »Algorithms are complex social and historical settings. changing the way in which criminal jus- Algorithms are human creations and as tice works through predictive policing, and increasingly such, subject to the same biases people re-shaping educational systems by incor- have. Its deployment depends, to a large porating automated evaluation tools. But being used extent, on the absorption of large stocks many of these developments have low lev- of data that can also be potentially biased The institution: els of transparency, public knowledge, and by governments (Lodge & Mennicken, 2017). Since much lack supervision mechanisms. The risks of of the processing, storage and use of in- this transition are also substantial, pos- and formation is performed by the algorithm ing severe governance challenges (Dafoe, itself and within a virtually inscrutable 2018). businesses.« black-box, experts are raising concerns In their 2019 Communique, G20 leaders as well about whether we can understand CIPPEC (Center for the Implementation of took a stand on this and stated the need to how this information is dealt with in order Public Policies for Equity and Growth) is an help societies adapt to the digital transfor- Principles are the starting point of a to scrutinize the decisions made and as- organization that strives to create an efficent, mation of our economies. They have also much more complex process and little sign both ethical and legal responsibility democratic, and just state to improve the qual- endorsed the OECD’s Artificial Intelligence progress has been made yet on the policy for the results reached (European Parlia- ity of life for all Argentine citizens. It focuses its efforts on analyzing and promoting public Principles, setting-up the ethics or base aspect of AI and the regulations that are ment Research Service, 2016). Information policies that encourage equity and growth in values for AI deployment, voicing their de- needed to reach those objectives. An ex- is power and the usage and recollection Argentina. sire to develop an AI centered on people, tensive and successful incorporation of AI of information without people’s express 84 85 GLOBAL SOLUTIONS JOURNAL ∙ ISSUE 5 THE FUTURE OF MULTILATERALISM: GLOBAL GOVERNANCE IN A CHANGING WORLD consent and knowledge infringes on their the need to develop shared standards and governance structures. While suprana- ministers approved these principles for AI rights (Kerry, 2019; Schrock, 2018). ensure interoperability, for example, of tional norms set common standards for as an annex to their declaration, which was As a result, several experts, sector privacy systems. Some of the proposals all countries that are part of these agree- later on ratified at the Osaka Summit. leaders and policymakers have agreed on include the creation of a guardian organ ments, and therefore ensure a baseline of Likewise, the European Union and the the need to act, and have launched a series of AI responsible for monitoring how al- rights and guarantees for their citizens, Nordic-Baltic region have also generated of guidelines for the set-up of AI systems. gorithms make decisions (Sample, 2017); such norms can overlook stark regional strategic plans for the development of But the debate about the best path for AI is a Council of National Robotics, without and local differences that may emerge not AI. From the EU perspective, it is not only not settled, the possible models are var- police power but with technical capacity only in terms of the countries’ technical about leading technological development ied, they cover a myriad of activities occur- to make recommendations (Calo, 2014); a capacities but also regarding their cul- but also leading on regulatory matters. ring across multiple jurisdictions, and few Federal Algorithm Directorate, modeled tural and political context. Simultaneously, The European Commission seeks to en- have been tested. The following lines pre- after the US’s FDA, with regulatory pow- while national and local norms can allow hance cooperation on AI across the EU to sent a conceptual approach for the existing ers to evaluate the systems before they for this diversity more easily, a fractional- boost its competitiveness and foster trust AI regulations, assessing the landscape of are launched on the market (Tutt, 2016); ized world in terms of people’s access to based on EU values and ethics. The logic technology governance across four differ- or a Machine Learning Commission that rights, for example, to privacy, can not only is human-centered and includes several ent dimensions regarding: (i) the regula- can create algorithms but without power reinforce present inequalities but can also requisites for AI systems to be considered tory lens through which technologies are to certify or approve these developments lead to more tension as a result of uneven reliable, regarding agency and human implicated, that is, the existence of verti- (Mulgan, 2016). technological developments. supervision; technical robustness, safety cal, sector-specific standards or transver- Many proposals have emerged from and the need for resilient systems to pre- sal regulations; (ii) their governance geo- international organizations in the past few vent or minimize unintentional damages; graphical scope: whether they are pushed » The EU seeks years, as geopolitical entities such as the privacy and data governance; transparen- forward by supranational organizations, UN, the EU and the OECD have begun to cy; diversity, non-discrimination and fair- countries or subnational governments; (iii) to enhance encourage the discussion on AI regula- ness; social and environmental well-be- the regulatory approach: whether more or tion. The goal behind many of these rec- ing; and accountability2. The Commission less coercive; and finally (iv) the public- cooperation ommendations is to generate a human- has presented these ethical guides to EU private divide in its making. In doing so, centered approach for the development of member states and different sector-spe- the brief’s goal is not to put forward all the on AI across AI, reducing differences among countries cific actors, setting-up a pilot phase with a regulations that exist but to discuss the and ensuring a minimum of guarantees High-Level Expert Group on AI, comprised possible alternatives currently under de- Europe to for all citizens. The OECD for instance, of 52 independent experts representing bate and challenges they bring about. has launched a Council on Artificial Intel- academia, industry and civil society, to The first dimension to assess regu- boost its ligence that published a series of general gather feedback. latory schemes has to do with the divide recommendations signed by 42 countries Moreover, the expert group also pre- between those who favor the sanction of competitiveness (36 belonging to the OECD, including the sented 33 recommendations to maximize vertical standards (eg. AI Sector Deal in US, Argentina, Brazil, Colombia, Costa AI’s impact on citizens, businesses, ad- the United Kingdom), and those who call and foster Rica, Perú and Romania)1. This document, ministrations and academia,